ABSTRACT
In this chapter, we discuss the optimality of empirical likelihood tests via their induced confidence regions. The optimality of parametric likelihood ratio tests is well-known. Although not formally stated, it is more or less implied that the optimality of likelihood ratio tests also carries over to empirical likelihood ratio tests. However, for right-censored data there are many variations of the empirical likelihood approach in the literature and obviously they are not all equivalent. If we are not careful, some variations of empirical likelihood tests may not be optimal. The induced Wilks confidence regions can be drastically different.
